Kalman-filter based stochastic-multiobjective network optimization and maximal-distance Latin hypercube sampling for uncertain inundation evacuation planning
Abstract. The subject of this research is to develop Kalman-filter based stochastic-multiobjective network optimization and maximal-distance Latin hypercube sampling methods regarding uncertain inundation evacuation planning. First, this research proposes a maximal-distance Latin hypercube sampling method to seek maximal space-filling sampling in uncertain flooding factor space. Uncertain inundation factors including upstream inflow, downstream water level, and channel friction resistance uncertainty are considered. Incorporated with the sampling method, HEC-RAS hydraulic model simulates stochastic flooding scenarios. Next, a Kalman-filter based stochastic-multiobjective network optimization model is established for uncertain inundation evacuation. Kalman-filter method iteratively predicts the flooding state of the next stage and updates prediction and decision according to new measurements. Kalman-filter based stochastic-multiobjective programming determines optimal shelter capacity expansion in the here-and-now stage and the best evacuation planning for each scenario in the wait-and-see stage. A case study of stochastic inundation evacuation in Muzha, Taiwan, is conducted. The contribution of this study is to incorporate Kalman-filter based stochastic-multiobjective network programming, HEC-RAS hydraulic simulation model, and maximal-distance Latin hypercube sampling to analyze inundation evacuation planning under uncertainty. The results show tradeoff between shelter expansion and evacuation time; furthermore, decreasing marginal effect of capacity expansion for evacuation time reduction is presented.